Very interesting question.
In my opinion, here's the way you need to think about it.
Let's assume for a second that feedback between the UE and the eNB is continuous, instantaneous and 100% accurate. In that case the scheduler always knows the optimal frequency resources to use to achieve the optimal link with the UE on both the uplink and the downlink. Since you can't do better than optimal, frequency hopping has no chance of improving capacity.
On the other hand, let's suppose you didn't have any feedback between the two, and your scheduler was memoryless, at least with regard to channel conditions. In that case, frequency hopping is the best you can do, because at least you're likely traversing some favorable frequencies some of the time, and if there's no feedback or memory, there's no opportunity to do better. The expression in English is that "you're hedging your bets".
So it seems to me that if you feel like you can't learn about the UE-eNB link fast enough, meaning you expect channel conditions to change significantly between the time the measurement is taken on one side of the link and applied at the other then frequency hop. If you're trying to get LTE out the door, and you feel you need to simplify the frequency selection component of the scheduler, frequency hop.
Otherwise, frequency hopping is not helping, moreover since you'll likely hop into bands that very poor for a given UE-to-eNB link, you are actually hurting your overall capacity.